Population viability analysis (PVA) is widely used in conservation biology to compare management strategies and predict probabilities of extinction for endangered species. However, it remains unclear whether the available range of generic PVA programs produce similar predictions when applied to common data sets. A comparative analysis was carried out using six commonly applied PVA packages (GAPPS, INMAT, RAMAS Age, RAMAS Metapop, RAMAS Stage and VORTEX), based on the extensive population data available for the endangered whooping crane, Grus americana. Significant differences were found between the projections of the PVA packages. In particular, the introduction of stochastic variation in breeding structure led to large differences between some packages. Surprisingly, two versions of VORTEX produced very different predictions due to a subtle difference in the way monogamous breeding was modelled. The pattern of similarities and differences among the projections of the packages differ depending on the species examined. In contrast to the predictions of comprehensive (realistic) models, simplified (standardized) models were found to be relatively congruent, which probably reflects similarities in the fundamental biological processes being modelled by the PVA packages. The implications for conservation are clear; caution must be exercised when interpreting the projections of a model built using any one PVA package since, due to seemingly subtle differences in what biological assumptions and threatening processes are included in the models, the predictions among PVA packages (or versions of the same program) are not necessarily concordant.